Today i’m running an open source program, which is ok under 0.3.0 but got error for 0.4.0 version. According to the error report, i found that new_dim and ctx.pad_tup in torch/nn/_functions/padding.py are all float type, which should be int type for successful running. So i’m a little confused that is there any difference in input parameters’ type of functional.Pad function? Thks!
They should still be integral types. Do you have a script that can reproduce the error?